In this study, the accuracy of LV volumes and LVEF derived from a single 15O-water PET/CT scan using cardiac-gated parametric blood volume images was assessed. High correlations between LV volumes and LVEF based on 15O-water parametric blood volume images towards MRI were found, despite the large number of steps required for our method. There were no significant differences between surface-, count- and volume-based methods. Agreement between PET and MRI was best for the volume-based method with no significant bias for ESV and LVEF, but with an overestimation of values for EDV and SV. For the surface-based method, a significant bias was found for ESV, SV and LVEF and no significant bias for EDV. The presence of one or two data points with high ESV, SV and LVEF show a relatively large positive difference between PET and MRI, giving the impression of a positive trend in the Bland Altman plots (Fig. 4). However, regression analysis of all Bland Altman plots gave a slope that was not significantly different from zero.
Input functions were based on cluster analysis of non-gated data, as the high statistical noise ruled out cluster analysis in single gated images. Four patients in this study had blood volume fractions of approximately 0.8 at the centre of the cavity, where it is supposed to be close to 1.0. When performing dynamic cardiac-gated acquisitions, the cardiac rebinning procedure excludes counts that originate from a cardiac cycle that ends up in between two frames of the dynamic scan. This will decrease the amplitude of the time-activity curves (TAC) in each voxel and result in an underestimation of the radioactivity concentrations especially during the first 5-s short frames, where one or two missed cycles will lead to a considerable reduction in counts compared to the non-gated input functions. In the modelling procedures, this will in turn lead to lowered blood volume fractions when cycles are missed during the first pass. Also, erroneously high PTF values will be assigned to voxels inside the cavity since the model will not be able to accurately describe the TAC of the left ventricle when the blood curve is underestimated during the first pass, which in turn will lead to erroneous estimation of volumes. This occurred in four patients in which the resulting volumetrics were generally poor when compared to MRI, and these patients were considered as outliers. This is a shortcoming of our current implementation of the cardiac rebinning procedure, and correction factors for the actual time contained in each frame should be addressed in future work. With the current implementation, it would be advisable to use a cutoff of the blood volume fraction (e.g. 0.9) below which LV volume and LVEF calculations should not be performed on parametric blood volume images. A possible modification of the data processing method that could avoid outliers as those found in the present work would be the definition of frames in terms of number of cardiac cycles instead of fixed durations in seconds. It was not possible to do this analysis with the current data, but this possibility will be investigated in future studies.
The use of gated first pass images is a more straightforward method compared to the construction of gated parametric V
B images. There are no modelling errors due to loss of counts from gated image reconstructions, and thus, higher correlations with MRI were found for FP images compared to for V
B images, for all parameters except for SV, when all patients were included. On the other hand, when outliers were excluded, V
B images showed higher correlations with MRI also for LVEF.
ECG-gated PET images are typically based on thresholding the inner contour of the tracer uptake in the myocardial wall, which is conceptually similar to the MRI approach. Myocardial wall uptake of tracers that are retained in proportion to MBF leads to variable cavity delineation and is known to produce errors in LV volumes and LVEF measurements in patients with chronic ischaemic heart disease [20]. Measuring cavity volumes using blood volume images effectively eliminates this error source, and equilibrium-gated blood volume imaging using planar scintigraphy or SPECT remains a clinically robust alternative to MRI [21]. The feasibility of synthetic blood volume images derived from parametric PET for LV volumes and LVEF measurements has not been shown before. However, true blood volume imaging by direct labelling of erythrocytes using inhaled 15O-CO for PET was shown to produce LV volumes and LVEF measurements with high accuracy [22, 23]. A recent publication [14] assessed the use of first pass 15O-water images, during the 15 s when the highest radioactivity concentrations were seen in the left ventricular cavity. This method showed a somewhat better correlation with MRI-based ESV, EDV, SV and LVEF values than the method suggested in our work. However, it should be noted that the range of volumes and LVEF in that study was much larger than in our data, which affects the correlation values. If only patients with MR-based LVEF > 53% were considered, as in our study, correlation using first pass images decreased to 0.40 for LVEF which is actually slightly lower than the present result.
The BPGS application showed some difficulties in the segmentation of parametric blood volume images, mostly when delineating the atrioventricular plane. Shrinking the edges of the volume for the automatic segmentation did help the system to delineate the atrioventricular plane better for some patients, but required manual adjustment, which might introduce observer bias. However, despite some difficulties in the segmentation, an excellent inter-operator repeatability was achieved for all parameters. Inter-operator repeatability was lowest for LVEF which partly could be explained by the narrow range of values.
The use of only eight gates is a drawback of this method, which tends to overestimate the end-systolic volumes in comparison with MRI. Using at least 16 gates is desirable; however, the low count statistics in the resulting images are likely to eliminate any potential benefit of using 16 gates instead of 8. An increase in injected dose might improve count statistics to a degree that would allow input curves to be more accurately derived from gated dynamic data. This would potentially decrease the risk of obtaining falsely low blood volume fractions and recover correct volumes from the outliers identified in this study. However, an injected activity of 400 MBq approaches the upper system limit regarding saturation. The use of the most recent generation of PET/CT or PET/MR scanners with a larger axial field of view and correspondingly higher sensitivity, time of flight capability and more robust counting statistics, might enable the use of a higher time resolution and higher doses.
In the present work, each image frame had to be sorted separately from the list-mode file, and then reconstructed into an eight-gate time-static image, involving the manual submission of 40 list-mode sorting or reconstruction assignments on the PET/CT reconstruction console and reconstruction of 160 image sets in total. Then, the 20 gated images were imported in a Matlab tool for resorting them into eight dynamic single cardiac-gate scans of 20 frames each, after which parametric images were calculated as described in the “Methods” section above. Aside from being very labour-intensive and time-consuming, this amount of manual processing is vulnerable to operator errors. Ideally, the scanner post-processing unit should automate the list-mode sorting, reconstruction and re-sorting into dynamic gated single cardiac-gated images, preferably using a frame timing definition that corresponds to full cardiac cycles. The software used for blood volume analysis originates from SPECT and has not been validated for synthetic PET blood volume images, but seems to provide reasonable results. If the software used for blood volume analysis would accommodate this, only end-systolic and end-diastolic single-gate dynamic images would need to be reconstructed. This would further reduce reconstruction times by 75%.
A limitation of the present work is that the patient population is limited to patients with mitral or aortic regurgitation with normal LV function. This highly specific patient population is though not likely to impact on the observed result. Segmentation of LV volumes is performed on either first pass images or parametric blood volume images, constructed from kinetic modelling, that are not affected by mitral or aortic regurgitation. Forward stroke volume or forward ejection fraction would have been affected but this was not a part of the present study. However, a more comprehensive assessment of the use of gated parametric 15O-water PET images in measuring LV volumes and LVEF is required in patients with reduced LV function. LVEF cutoff used for a decision on defibrillator implants is 35%, which is lower than in the present study, and another study will have to be done to qualify 15O-water in that range.